net/openvswitch/flow.c

   1 /*
   2  * Copyright (c) 2007-2011 Nicira, Inc.
   3  *
   4  * This program is free software; you can redistribute it and/or
   5  * modify it under the terms of version 2 of the GNU General Public
   6  * License as published by the Free Software Foundation.
   7  *
   8  * This program is distributed in the hope that it will be useful, but
   9  * WITHOUT ANY WARRANTY; without even the implied warranty of
  10  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
  11  * General Public License for more details.
  12  *
  13  * You should have received a copy of the GNU General Public License
  14  * along with this program; if not, write to the Free Software
  15  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA
  16  * 02110-1301, USA
  17  */
  18
  19 #include "flow.h"
  20 #include "datapath.h"
  21 #include <linux/uaccess.h>
  22 #include <linux/netdevice.h>
  23 #include <linux/etherdevice.h>
  24 #include <linux/if_ether.h>
  25 #include <linux/if_vlan.h>
  26 #include <net/llc_pdu.h>
  27 #include <linux/kernel.h>
  28 #include <linux/jhash.h>
  29 #include <linux/jiffies.h>
  30 #include <linux/llc.h>
  31 #include <linux/module.h>
  32 #include <linux/in.h>
  33 #include <linux/rcupdate.h>
  34 #include <linux/if_arp.h>
  35 #include <linux/ip.h>
  36 #include <linux/ipv6.h>
  37 #include <linux/tcp.h>
  38 #include <linux/udp.h>
  39 #include <linux/icmp.h>
  40 #include <linux/icmpv6.h>
  41 #include <linux/rculist.h>
  42 #include <net/ip.h>
  43 #include <net/ipv6.h>
  44 #include <net/ndisc.h>
  45
  46 static struct kmem_cache *flow_cache;
  47
  48 static int check_header(struct sk_buff *skb, int len)
  49 {
  50         if (unlikely(skb->len < len))
  51                 return -EINVAL;
  52         if (unlikely(!pskb_may_pull(skb, len)))
  53                 return -ENOMEM;
  54         return 0;
  55 }
  56
  57 static bool arphdr_ok(struct sk_buff *skb)
  58 {
  59         return pskb_may_pull(skb, skb_network_offset(skb) +
  60                                   sizeof(struct arp_eth_header));
  61 }
  62
  63 static int check_iphdr(struct sk_buff *skb)
  64 {
  65         unsigned int nh_ofs = skb_network_offset(skb);
  66         unsigned int ip_len;
  67         int err;
  68
  69         err = check_header(skb, nh_ofs + sizeof(struct iphdr));
  70         if (unlikely(err))
  71                 return err;
  72
  73         ip_len = ip_hdrlen(skb);
  74         if (unlikely(ip_len < sizeof(struct iphdr) ||
  75                      skb->len < nh_ofs + ip_len))
  76                 return -EINVAL;
  77
  78         skb_set_transport_header(skb, nh_ofs + ip_len);
  79         return 0;
  80 }
  81
  82 static bool tcphdr_ok(struct sk_buff *skb)
  83 {
  84         int th_ofs = skb_transport_offset(skb);
  85         int tcp_len;
  86
  87         if (unlikely(!pskb_may_pull(skb, th_ofs + sizeof(struct tcphdr))))
  88                 return false;
  89
  90         tcp_len = tcp_hdrlen(skb);
  91         if (unlikely(tcp_len < sizeof(struct tcphdr) ||
  92                      skb->len < th_ofs + tcp_len))
  93                 return false;
  94
  95         return true;
  96 }
  97
  98 static bool udphdr_ok(struct sk_buff *skb)
  99 {
 100         return pskb_may_pull(skb, skb_transport_offset(skb) +
 101                                   sizeof(struct udphdr));
 102 }
 103
 104 static bool icmphdr_ok(struct sk_buff *skb)
 105 {
 106         return pskb_may_pull(skb, skb_transport_offset(skb) +
 107                                   sizeof(struct icmphdr));
 108 }
 109
 110 u64 ovs_flow_used_time(unsigned long flow_jiffies)
 111 {
 112         struct timespec cur_ts;
 113         u64 cur_ms, idle_ms;
 114
 115         ktime_get_ts(&cur_ts);
 116         idle_ms = jiffies_to_msecs(jiffies - flow_jiffies);
 117         cur_ms = (u64)cur_ts.tv_sec * MSEC_PER_SEC +
 118                  cur_ts.tv_nsec / NSEC_PER_MSEC;
 119
 120         return cur_ms - idle_ms;
 121 }
 122
 123 #define SW_FLOW_KEY_OFFSET(field)               \
 124         (offsetof(struct sw_flow_key, field) +  \
 125          FIELD_SIZEOF(struct sw_flow_key, field))
 126
 127 static int parse_ipv6hdr(struct sk_buff *skb, struct sw_flow_key *key,
 128                          int *key_lenp)
 129 {
 130         unsigned int nh_ofs = skb_network_offset(skb);
 131         unsigned int nh_len;
 132         int payload_ofs;
 133         struct ipv6hdr *nh;
 134         uint8_t nexthdr;
 135         __be16 frag_off;
 136         int err;
 137
 138         *key_lenp = SW_FLOW_KEY_OFFSET(ipv6.label);
 139
 140         err = check_header(skb, nh_ofs + sizeof(*nh));
 141         if (unlikely(err))
 142                 return err;
 143
 144         nh = ipv6_hdr(skb);
 145         nexthdr = nh->nexthdr;
 146         payload_ofs = (u8 *)(nh + 1) - skb->data;
 147
 148         key->ip.proto = NEXTHDR_NONE;
 149         key->ip.tos = ipv6_get_dsfield(nh);
 150         key->ip.ttl = nh->hop_limit;
 151         key->ipv6.label = *(__be32 *)nh & htonl(IPV6_FLOWINFO_FLOWLABEL);
 152         key->ipv6.addr.src = nh->saddr;
 153         key->ipv6.addr.dst = nh->daddr;
 154
 155         payload_ofs = ipv6_skip_exthdr(skb, payload_ofs, &nexthdr, &frag_off);
 156         if (unlikely(payload_ofs < 0))
 157                 return -EINVAL;
 158
 159         if (frag_off) {
 160                 if (frag_off & htons(~0x7))
 161                         key->ip.frag = OVS_FRAG_TYPE_LATER;
 162                 else
 163                         key->ip.frag = OVS_FRAG_TYPE_FIRST;
 164         }
 165
 166         nh_len = payload_ofs - nh_ofs;
 167         skb_set_transport_header(skb, nh_ofs + nh_len);
 168         key->ip.proto = nexthdr;
 169         return nh_len;
 170 }
 171
 172 static bool icmp6hdr_ok(struct sk_buff *skb)
 173 {
 174         return pskb_may_pull(skb, skb_transport_offset(skb) +
 175                                   sizeof(struct icmp6hdr));
 176 }
 177
 178 #define TCP_FLAGS_OFFSET 13
 179 #define TCP_FLAG_MASK 0x3f
 180
 181 void ovs_flow_used(struct sw_flow *flow, struct sk_buff *skb)
 182 {
 183         u8 tcp_flags = 0;
 184
 185         if ((flow->key.eth.type == htons(ETH_P_IP) ||
 186              flow->key.eth.type == htons(ETH_P_IPV6)) &&
 187             flow->key.ip.proto == IPPROTO_TCP &&
 188             likely(skb->len >= skb_transport_offset(skb) + sizeof(struct tcphdr))) {
 189                 u8 *tcp = (u8 *)tcp_hdr(skb);
 190                 tcp_flags = *(tcp + TCP_FLAGS_OFFSET) & TCP_FLAG_MASK;
 191         }
 192
 193         spin_lock(&flow->lock);
 194         flow->used = jiffies;
 195         flow->packet_count++;
 196         flow->byte_count += skb->len;
 197         flow->tcp_flags |= tcp_flags;
 198         spin_unlock(&flow->lock);
 199 }
 200
 201 struct sw_flow_actions *ovs_flow_actions_alloc(const struct nlattr *actions)
 202 {
 203         int actions_len = nla_len(actions);
 204         struct sw_flow_actions *sfa;
 205
 206         /* At least DP_MAX_PORTS actions are required to be able to flood a
 207          * packet to every port.  Factor of 2 allows for setting VLAN tags,
 208          * etc. */
 209         if (actions_len > 2 * DP_MAX_PORTS * nla_total_size(4))
 210                 return ERR_PTR(-EINVAL);
 211
 212         sfa = kmalloc(sizeof(*sfa) + actions_len, GFP_KERNEL);
 213         if (!sfa)
 214                 return ERR_PTR(-ENOMEM);
 215
 216         sfa->actions_len = actions_len;
 217         memcpy(sfa->actions, nla_data(actions), actions_len);
 218         return sfa;
 219 }
 220
 221 struct sw_flow *ovs_flow_alloc(void)
 222 {
 223         struct sw_flow *flow;
 224
 225         flow = kmem_cache_alloc(flow_cache, GFP_KERNEL);
 226         if (!flow)
 227                 return ERR_PTR(-ENOMEM);
 228
 229         spin_lock_init(&flow->lock);
 230         flow->sf_acts = NULL;
 231
 232         return flow;
 233 }
 234
 235 static struct hlist_head *find_bucket(struct flow_table *table, u32 hash)
 236 {
 237         hash = jhash_1word(hash, table->hash_seed);
 238         return flex_array_get(table->buckets,
 239                                 (hash & (table->n_buckets - 1)));
 240 }
 241
 242 static struct flex_array *alloc_buckets(unsigned int n_buckets)
 243 {
 244         struct flex_array *buckets;
 245         int i, err;
 246
 247         buckets = flex_array_alloc(sizeof(struct hlist_head *),
 248                                    n_buckets, GFP_KERNEL);
 249         if (!buckets)
 250                 return NULL;
 251
 252         err = flex_array_prealloc(buckets, 0, n_buckets, GFP_KERNEL);
 253         if (err) {
 254                 flex_array_free(buckets);
 255                 return NULL;
 256         }
 257
 258         for (i = 0; i < n_buckets; i++)
 259                 INIT_HLIST_HEAD((struct hlist_head *)
 260                                         flex_array_get(buckets, i));
 261
 262         return buckets;
 263 }
 264
 265 static void free_buckets(struct flex_array *buckets)
 266 {
 267         flex_array_free(buckets);
 268 }
 269
 270 struct flow_table *ovs_flow_tbl_alloc(int new_size)
 271 {
 272         struct flow_table *table = kmalloc(sizeof(*table), GFP_KERNEL);
 273
 274         if (!table)
 275                 return NULL;
 276
 277         table->buckets = alloc_buckets(new_size);
 278
 279         if (!table->buckets) {
 280                 kfree(table);
 281                 return NULL;
 282         }
 283         table->n_buckets = new_size;
 284         table->count = 0;
 285         table->node_ver = 0;
 286         table->keep_flows = false;
 287         get_random_bytes(&table->hash_seed, sizeof(u32));
 288
 289         return table;
 290 }
 291
 292 void ovs_flow_tbl_destroy(struct flow_table *table)
 293 {
 294         int i;
 295
 296         if (!table)
 297                 return;
 298
 299         if (table->keep_flows)
 300                 goto skip_flows;
 301
 302         for (i = 0; i < table->n_buckets; i++) {
 303                 struct sw_flow *flow;
 304                 struct hlist_head *head = flex_array_get(table->buckets, i);
 305                 struct hlist_node *node, *n;
 306                 int ver = table->node_ver;
 307
 308                 hlist_for_each_entry_safe(flow, node, n, head, hash_node[ver]) {
 309                         hlist_del_rcu(&flow->hash_node[ver]);
 310                         ovs_flow_free(flow);
 311                 }
 312         }
 313
 314 skip_flows:
 315         free_buckets(table->buckets);
 316         kfree(table);
 317 }
 318
 319 static void flow_tbl_destroy_rcu_cb(struct rcu_head *rcu)
 320 {
 321         struct flow_table *table = container_of(rcu, struct flow_table, rcu);
 322
 323         ovs_flow_tbl_destroy(table);
 324 }
 325
 326 void ovs_flow_tbl_deferred_destroy(struct flow_table *table)
 327 {
 328         if (!table)
 329                 return;
 330
 331         call_rcu(&table->rcu, flow_tbl_destroy_rcu_cb);
 332 }
 333
 334 struct sw_flow *ovs_flow_tbl_next(struct flow_table *table, u32 *bucket, u32 *last)
 335 {
 336         struct sw_flow *flow;
 337         struct hlist_head *head;
 338         struct hlist_node *n;
 339         int ver;
 340         int i;
 341
 342         ver = table->node_ver;
 343         while (*bucket < table->n_buckets) {
 344                 i = 0;
 345                 head = flex_array_get(table->buckets, *bucket);
 346                 hlist_for_each_entry_rcu(flow, n, head, hash_node[ver]) {
 347                         if (i < *last) {
 348                                 i++;
 349                                 continue;
 350                         }
 351                         *last = i + 1;
 352                         return flow;
 353                 }
 354                 (*bucket)++;
 355                 *last = 0;
 356         }
 357
 358         return NULL;
 359 }
 360
 361 static void flow_table_copy_flows(struct flow_table *old, struct flow_table *new)
 362 {
 363         int old_ver;
 364         int i;
 365
 366         old_ver = old->node_ver;
 367         new->node_ver = !old_ver;
 368
 369         /* Insert in new table. */
 370         for (i = 0; i < old->n_buckets; i++) {
 371                 struct sw_flow *flow;
 372                 struct hlist_head *head;
 373                 struct hlist_node *n;
 374
 375                 head = flex_array_get(old->buckets, i);
 376
 377                 hlist_for_each_entry(flow, n, head, hash_node[old_ver])
 378                         ovs_flow_tbl_insert(new, flow);
 379         }
 380         old->keep_flows = true;
 381 }
 382
 383 static struct flow_table *__flow_tbl_rehash(struct flow_table *table, int n_buckets)
 384 {
 385         struct flow_table *new_table;
 386
 387         new_table = ovs_flow_tbl_alloc(n_buckets);
 388         if (!new_table)
 389                 return ERR_PTR(-ENOMEM);
 390
 391         flow_table_copy_flows(table, new_table);
 392
 393         return new_table;
 394 }
 395
 396 struct flow_table *ovs_flow_tbl_rehash(struct flow_table *table)
 397 {
 398         return __flow_tbl_rehash(table, table->n_buckets);
 399 }
 400
 401 struct flow_table *ovs_flow_tbl_expand(struct flow_table *table)
 402 {
 403         return __flow_tbl_rehash(table, table->n_buckets * 2);
 404 }
 405
 406 void ovs_flow_free(struct sw_flow *flow)
 407 {
 408         if (unlikely(!flow))
 409                 return;
 410
 411         kfree((struct sf_flow_acts __force *)flow->sf_acts);
 412         kmem_cache_free(flow_cache, flow);
 413 }
 414
 415 /* RCU callback used by ovs_flow_deferred_free. */
 416 static void rcu_free_flow_callback(struct rcu_head *rcu)
 417 {
 418         struct sw_flow *flow = container_of(rcu, struct sw_flow, rcu);
 419
 420         ovs_flow_free(flow);
 421 }
 422
 423 /* Schedules 'flow' to be freed after the next RCU grace period.
 424  * The caller must hold rcu_read_lock for this to be sensible. */
 425 void ovs_flow_deferred_free(struct sw_flow *flow)
 426 {
 427         call_rcu(&flow->rcu, rcu_free_flow_callback);
 428 }
 429
 430 /* RCU callback used by ovs_flow_deferred_free_acts. */
 431 static void rcu_free_acts_callback(struct rcu_head *rcu)
 432 {
 433         struct sw_flow_actions *sf_acts = container_of(rcu,
 434                         struct sw_flow_actions, rcu);
 435         kfree(sf_acts);
 436 }
 437
 438 /* Schedules 'sf_acts' to be freed after the next RCU grace period.
 439  * The caller must hold rcu_read_lock for this to be sensible. */
 440 void ovs_flow_deferred_free_acts(struct sw_flow_actions *sf_acts)
 441 {
 442         call_rcu(&sf_acts->rcu, rcu_free_acts_callback);
 443 }
 444
 445 static int parse_vlan(struct sk_buff *skb, struct sw_flow_key *key)
 446 {
 447         struct qtag_prefix {
 448                 __be16 eth_type; /* ETH_P_8021Q */
 449                 __be16 tci;
 450         };
 451         struct qtag_prefix *qp;
 452
 453         if (unlikely(skb->len < sizeof(struct qtag_prefix) + sizeof(__be16)))
 454                 return 0;
 455
 456         if (unlikely(!pskb_may_pull(skb, sizeof(struct qtag_prefix) +
 457                                          sizeof(__be16))))
 458                 return -ENOMEM;
 459
 460         qp = (struct qtag_prefix *) skb->data;
 461         key->eth.tci = qp->tci | htons(VLAN_TAG_PRESENT);
 462         __skb_pull(skb, sizeof(struct qtag_prefix));
 463
 464         return 0;
 465 }
 466
 467 static __be16 parse_ethertype(struct sk_buff *skb)
 468 {
 469         struct llc_snap_hdr {
 470                 u8  dsap;  /* Always 0xAA */
 471                 u8  ssap;  /* Always 0xAA */
 472                 u8  ctrl;
 473                 u8  oui[3];
 474                 __be16 ethertype;
 475         };
 476         struct llc_snap_hdr *llc;
 477         __be16 proto;
 478
 479         proto = *(__be16 *) skb->data;
 480         __skb_pull(skb, sizeof(__be16));
 481
 482         if (ntohs(proto) >= 1536)
 483                 return proto;
 484
 485         if (skb->len < sizeof(struct llc_snap_hdr))
 486                 return htons(ETH_P_802_2);
 487
 488         if (unlikely(!pskb_may_pull(skb, sizeof(struct llc_snap_hdr))))
 489                 return htons(0);
 490
 491         llc = (struct llc_snap_hdr *) skb->data;
 492         if (llc->dsap != LLC_SAP_SNAP ||
 493             llc->ssap != LLC_SAP_SNAP ||
 494             (llc->oui[0] | llc->oui[1] | llc->oui[2]) != 0)
 495                 return htons(ETH_P_802_2);
 496
 497         __skb_pull(skb, sizeof(struct llc_snap_hdr));
 498         return llc->ethertype;
 499 }
 500
 501 static int parse_icmpv6(struct sk_buff *skb, struct sw_flow_key *key,
 502                         int *key_lenp, int nh_len)
 503 {
 504         struct icmp6hdr *icmp = icmp6_hdr(skb);
 505         int error = 0;
 506         int key_len;
 507
 508         /* The ICMPv6 type and code fields use the 16-bit transport port
 509          * fields, so we need to store them in 16-bit network byte order.
 510          */
 511         key->ipv6.tp.src = htons(icmp->icmp6_type);
 512         key->ipv6.tp.dst = htons(icmp->icmp6_code);
 513         key_len = SW_FLOW_KEY_OFFSET(ipv6.tp);
 514
 515         if (icmp->icmp6_code == 0 &&
 516             (icmp->icmp6_type == NDISC_NEIGHBOUR_SOLICITATION ||
 517              icmp->icmp6_type == NDISC_NEIGHBOUR_ADVERTISEMENT)) {
 518                 int icmp_len = skb->len - skb_transport_offset(skb);
 519                 struct nd_msg *nd;
 520                 int offset;
 521
 522                 key_len = SW_FLOW_KEY_OFFSET(ipv6.nd);
 523
 524                 /* In order to process neighbor discovery options, we need the
 525                  * entire packet.
 526                  */
 527                 if (unlikely(icmp_len < sizeof(*nd)))
 528                         goto out;
 529                 if (unlikely(skb_linearize(skb))) {
 530                         error = -ENOMEM;
 531                         goto out;
 532                 }
 533
 534                 nd = (struct nd_msg *)skb_transport_header(skb);
 535                 key->ipv6.nd.target = nd->target;
 536                 key_len = SW_FLOW_KEY_OFFSET(ipv6.nd);
 537
 538                 icmp_len -= sizeof(*nd);
 539                 offset = 0;
 540                 while (icmp_len >= 8) {
 541                         struct nd_opt_hdr *nd_opt =
 542                                  (struct nd_opt_hdr *)(nd->opt + offset);
 543                         int opt_len = nd_opt->nd_opt_len * 8;
 544
 545                         if (unlikely(!opt_len || opt_len > icmp_len))
 546                                 goto invalid;
 547
 548                         /* Store the link layer address if the appropriate
 549                          * option is provided.  It is considered an error if
 550                          * the same link layer option is specified twice.
 551                          */
 552                         if (nd_opt->nd_opt_type == ND_OPT_SOURCE_LL_ADDR
 553                             && opt_len == 8) {
 554                                 if (unlikely(!is_zero_ether_addr(key->ipv6.nd.sll)))
 555                                         goto invalid;
 556                                 memcpy(key->ipv6.nd.sll,
 557                                     &nd->opt[offset+sizeof(*nd_opt)], ETH_ALEN);
 558                         } else if (nd_opt->nd_opt_type == ND_OPT_TARGET_LL_ADDR
 559                                    && opt_len == 8) {
 560                                 if (unlikely(!is_zero_ether_addr(key->ipv6.nd.tll)))
 561                                         goto invalid;
 562                                 memcpy(key->ipv6.nd.tll,
 563                                     &nd->opt[offset+sizeof(*nd_opt)], ETH_ALEN);
 564                         }
 565
 566                         icmp_len -= opt_len;
 567                         offset += opt_len;
 568                 }
 569         }
 570
 571         goto out;
 572
 573 invalid:
 574         memset(&key->ipv6.nd.target, 0, sizeof(key->ipv6.nd.target));
 575         memset(key->ipv6.nd.sll, 0, sizeof(key->ipv6.nd.sll));
 576         memset(key->ipv6.nd.tll, 0, sizeof(key->ipv6.nd.tll));
 577
 578 out:
 579         *key_lenp = key_len;
 580         return error;
 581 }
 582
 583 /**
 584  * ovs_flow_extract - extracts a flow key from an Ethernet frame.
 585  * @skb: sk_buff that contains the frame, with skb->data pointing to the
 586  * Ethernet header
 587  * @in_port: port number on which @skb was received.
 588  * @key: output flow key
 589  * @key_lenp: length of output flow key
 590  *
 591  * The caller must ensure that skb->len >= ETH_HLEN.
 592  *
 593  * Returns 0 if successful, otherwise a negative errno value.
 594  *
 595  * Initializes @skb header pointers as follows:
 596  *
 597  *    - skb->mac_header: the Ethernet header.
 598  *
 599  *    - skb->network_header: just past the Ethernet header, or just past the
 600  *      VLAN header, to the first byte of the Ethernet payload.
 601  *
 602  *    - skb->transport_header: If key->dl_type is ETH_P_IP or ETH_P_IPV6
 603  *      on output, then just past the IP header, if one is present and
 604  *      of a correct length, otherwise the same as skb->network_header.
 605  *      For other key->dl_type values it is left untouched.
 606  */
 607 int ovs_flow_extract(struct sk_buff *skb, u16 in_port, struct sw_flow_key *key,
 608                  int *key_lenp)
 609 {
 610         int error = 0;
 611         int key_len = SW_FLOW_KEY_OFFSET(eth);
 612         struct ethhdr *eth;
 613
 614         memset(key, 0, sizeof(*key));
 615
 616         key->phy.priority = skb->priority;
 617         key->phy.in_port = in_port;
 618
 619         skb_reset_mac_header(skb);
 620
 621         /* Link layer.  We are guaranteed to have at least the 14 byte Ethernet
 622          * header in the linear data area.
 623          */
 624         eth = eth_hdr(skb);
 625         memcpy(key->eth.src, eth->h_source, ETH_ALEN);
 626         memcpy(key->eth.dst, eth->h_dest, ETH_ALEN);
 627
 628         __skb_pull(skb, 2 * ETH_ALEN);
 629
 630         if (vlan_tx_tag_present(skb))
 631                 key->eth.tci = htons(skb->vlan_tci);
 632         else if (eth->h_proto == htons(ETH_P_8021Q))
 633                 if (unlikely(parse_vlan(skb, key)))
 634                         return -ENOMEM;
 635
 636         key->eth.type = parse_ethertype(skb);
 637         if (unlikely(key->eth.type == htons(0)))
 638                 return -ENOMEM;
 639
 640         skb_reset_network_header(skb);
 641         __skb_push(skb, skb->data - skb_mac_header(skb));
 642
 643         /* Network layer. */
 644         if (key->eth.type == htons(ETH_P_IP)) {
 645                 struct iphdr *nh;
 646                 __be16 offset;
 647
 648                 key_len = SW_FLOW_KEY_OFFSET(ipv4.addr);
 649
 650                 error = check_iphdr(skb);
 651                 if (unlikely(error)) {
 652                         if (error == -EINVAL) {
 653                                 skb->transport_header = skb->network_header;
 654                                 error = 0;
 655                         }
 656                         goto out;
 657                 }
 658
 659                 nh = ip_hdr(skb);
 660                 key->ipv4.addr.src = nh->saddr;
 661                 key->ipv4.addr.dst = nh->daddr;
 662
 663                 key->ip.proto = nh->protocol;
 664                 key->ip.tos = nh->tos;
 665                 key->ip.ttl = nh->ttl;
 666
 667                 offset = nh->frag_off & htons(IP_OFFSET);
 668                 if (offset) {
 669                         key->ip.frag = OVS_FRAG_TYPE_LATER;
 670                         goto out;
 671                 }
 672                 if (nh->frag_off & htons(IP_MF) ||
 673                          skb_shinfo(skb)->gso_type & SKB_GSO_UDP)
 674                         key->ip.frag = OVS_FRAG_TYPE_FIRST;
 675
 676                 /* Transport layer. */
 677                 if (key->ip.proto == IPPROTO_TCP) {
 678                         key_len = SW_FLOW_KEY_OFFSET(ipv4.tp);
 679                         if (tcphdr_ok(skb)) {
 680                                 struct tcphdr *tcp = tcp_hdr(skb);
 681                                 key->ipv4.tp.src = tcp->source;
 682                                 key->ipv4.tp.dst = tcp->dest;
 683                         }
 684                 } else if (key->ip.proto == IPPROTO_UDP) {
 685                         key_len = SW_FLOW_KEY_OFFSET(ipv4.tp);
 686                         if (udphdr_ok(skb)) {
 687                                 struct udphdr *udp = udp_hdr(skb);
 688                                 key->ipv4.tp.src = udp->source;
 689                                 key->ipv4.tp.dst = udp->dest;
 690                         }
 691                 } else if (key->ip.proto == IPPROTO_ICMP) {
 692                         key_len = SW_FLOW_KEY_OFFSET(ipv4.tp);
 693                         if (icmphdr_ok(skb)) {
 694                                 struct icmphdr *icmp = icmp_hdr(skb);
 695                                 /* The ICMP type and code fields use the 16-bit
 696                                  * transport port fields, so we need to store
 697                                  * them in 16-bit network byte order. */
 698                                 key->ipv4.tp.src = htons(icmp->type);
 699                                 key->ipv4.tp.dst = htons(icmp->code);
 700                         }
 701                 }
 702
 703         } else if (key->eth.type == htons(ETH_P_ARP) && arphdr_ok(skb)) {
 704                 struct arp_eth_header *arp;
 705
 706                 arp = (struct arp_eth_header *)skb_network_header(skb);
 707
 708                 if (arp->ar_hrd == htons(ARPHRD_ETHER)
 709                                 && arp->ar_pro == htons(ETH_P_IP)
 710                                 && arp->ar_hln == ETH_ALEN
 711                                 && arp->ar_pln == 4) {
 712
 713                         /* We only match on the lower 8 bits of the opcode. */
 714                         if (ntohs(arp->ar_op) <= 0xff)
 715                                 key->ip.proto = ntohs(arp->ar_op);
 716
 717                         if (key->ip.proto == ARPOP_REQUEST
 718                                         || key->ip.proto == ARPOP_REPLY) {
 719                                 memcpy(&key->ipv4.addr.src, arp->ar_sip, sizeof(key->ipv4.addr.src));
 720                                 memcpy(&key->ipv4.addr.dst, arp->ar_tip, sizeof(key->ipv4.addr.dst));
 721                                 memcpy(key->ipv4.arp.sha, arp->ar_sha, ETH_ALEN);
 722                                 memcpy(key->ipv4.arp.tha, arp->ar_tha, ETH_ALEN);
 723                                 key_len = SW_FLOW_KEY_OFFSET(ipv4.arp);
 724                         }
 725                 }
 726         } else if (key->eth.type == htons(ETH_P_IPV6)) {
 727                 int nh_len;             /* IPv6 Header + Extensions */
 728
 729                 nh_len = parse_ipv6hdr(skb, key, &key_len);
 730                 if (unlikely(nh_len < 0)) {
 731                         if (nh_len == -EINVAL)
 732                                 skb->transport_header = skb->network_header;
 733                         else
 734                                 error = nh_len;
 735                         goto out;
 736                 }
 737
 738                 if (key->ip.frag == OVS_FRAG_TYPE_LATER)
 739                         goto out;
 740                 if (skb_shinfo(skb)->gso_type & SKB_GSO_UDP)
 741                         key->ip.frag = OVS_FRAG_TYPE_FIRST;
 742
 743                 /* Transport layer. */
 744                 if (key->ip.proto == NEXTHDR_TCP) {
 745                         key_len = SW_FLOW_KEY_OFFSET(ipv6.tp);
 746                         if (tcphdr_ok(skb)) {
 747                                 struct tcphdr *tcp = tcp_hdr(skb);
 748                                 key->ipv6.tp.src = tcp->source;
 749                                 key->ipv6.tp.dst = tcp->dest;
 750                         }
 751                 } else if (key->ip.proto == NEXTHDR_UDP) {
 752                         key_len = SW_FLOW_KEY_OFFSET(ipv6.tp);
 753                         if (udphdr_ok(skb)) {
 754                                 struct udphdr *udp = udp_hdr(skb);
 755                                 key->ipv6.tp.src = udp->source;
 756                                 key->ipv6.tp.dst = udp->dest;
 757                         }
 758                 } else if (key->ip.proto == NEXTHDR_ICMP) {
 759                         key_len = SW_FLOW_KEY_OFFSET(ipv6.tp);
 760                         if (icmp6hdr_ok(skb)) {
 761                                 error = parse_icmpv6(skb, key, &key_len, nh_len);
 762                                 if (error < 0)
 763                                         goto out;
 764                         }
 765                 }
 766         }
 767
 768 out:
 769         *key_lenp = key_len;
 770         return error;
 771 }
 772
 773 u32 ovs_flow_hash(const struct sw_flow_key *key, int key_len)
 774 {
 775         return jhash2((u32 *)key, DIV_ROUND_UP(key_len, sizeof(u32)), 0);
 776 }
 777
 778 struct sw_flow *ovs_flow_tbl_lookup(struct flow_table *table,
 779                                 struct sw_flow_key *key, int key_len)
 780 {
 781         struct sw_flow *flow;
 782         struct hlist_node *n;
 783         struct hlist_head *head;
 784         u32 hash;
 785
 786         hash = ovs_flow_hash(key, key_len);
 787
 788         head = find_bucket(table, hash);
 789         hlist_for_each_entry_rcu(flow, n, head, hash_node[table->node_ver]) {
 790
 791                 if (flow->hash == hash &&
 792                     !memcmp(&flow->key, key, key_len)) {
 793                         return flow;
 794                 }
 795         }
 796         return NULL;
 797 }
 798
 799 void ovs_flow_tbl_insert(struct flow_table *table, struct sw_flow *flow)
 800 {
 801         struct hlist_head *head;
 802
 803         head = find_bucket(table, flow->hash);
 804         hlist_add_head_rcu(&flow->hash_node[table->node_ver], head);
 805         table->count++;
 806 }
 807
 808 void ovs_flow_tbl_remove(struct flow_table *table, struct sw_flow *flow)
 809 {
 810         hlist_del_rcu(&flow->hash_node[table->node_ver]);
 811         table->count--;
 812         BUG_ON(table->count < 0);
 813 }
 814
 815 /* The size of the argument for each %OVS_KEY_ATTR_* Netlink attribute.  */
 816 const int ovs_key_lens[OVS_KEY_ATTR_MAX + 1] = {
 817         [OVS_KEY_ATTR_ENCAP] = -1,
 818         [OVS_KEY_ATTR_PRIORITY] = sizeof(u32),
 819         [OVS_KEY_ATTR_IN_PORT] = sizeof(u32),
 820         [OVS_KEY_ATTR_ETHERNET] = sizeof(struct ovs_key_ethernet),
 821         [OVS_KEY_ATTR_VLAN] = sizeof(__be16),
 822         [OVS_KEY_ATTR_ETHERTYPE] = sizeof(__be16),
 823         [OVS_KEY_ATTR_IPV4] = sizeof(struct ovs_key_ipv4),
 824         [OVS_KEY_ATTR_IPV6] = sizeof(struct ovs_key_ipv6),
 825         [OVS_KEY_ATTR_TCP] = sizeof(struct ovs_key_tcp),
 826         [OVS_KEY_ATTR_UDP] = sizeof(struct ovs_key_udp),
 827         [OVS_KEY_ATTR_ICMP] = sizeof(struct ovs_key_icmp),
 828         [OVS_KEY_ATTR_ICMPV6] = sizeof(struct ovs_key_icmpv6),
 829         [OVS_KEY_ATTR_ARP] = sizeof(struct ovs_key_arp),
 830         [OVS_KEY_ATTR_ND] = sizeof(struct ovs_key_nd),
 831 };
 832
 833 static int ipv4_flow_from_nlattrs(struct sw_flow_key *swkey, int *key_len,
 834                                   const struct nlattr *a[], u32 *attrs)
 835 {
 836         const struct ovs_key_icmp *icmp_key;
 837         const struct ovs_key_tcp *tcp_key;
 838         const struct ovs_key_udp *udp_key;
 839
 840         switch (swkey->ip.proto) {
 841         case IPPROTO_TCP:
 842                 if (!(*attrs & (1 << OVS_KEY_ATTR_TCP)))
 843                         return -EINVAL;
 844                 *attrs &= ~(1 << OVS_KEY_ATTR_TCP);
 845
 846                 *key_len = SW_FLOW_KEY_OFFSET(ipv4.tp);
 847                 tcp_key = nla_data(a[OVS_KEY_ATTR_TCP]);
 848                 swkey->ipv4.tp.src = tcp_key->tcp_src;
 849                 swkey->ipv4.tp.dst = tcp_key->tcp_dst;
 850                 break;
 851
 852         case IPPROTO_UDP:
 853                 if (!(*attrs & (1 << OVS_KEY_ATTR_UDP)))
 854                         return -EINVAL;
 855                 *attrs &= ~(1 << OVS_KEY_ATTR_UDP);
 856
 857                 *key_len = SW_FLOW_KEY_OFFSET(ipv4.tp);
 858                 udp_key = nla_data(a[OVS_KEY_ATTR_UDP]);
 859                 swkey->ipv4.tp.src = udp_key->udp_src;
 860                 swkey->ipv4.tp.dst = udp_key->udp_dst;
 861                 break;
 862
 863         case IPPROTO_ICMP:
 864                 if (!(*attrs & (1 << OVS_KEY_ATTR_ICMP)))
 865                         return -EINVAL;
 866                 *attrs &= ~(1 << OVS_KEY_ATTR_ICMP);
 867
 868                 *key_len = SW_FLOW_KEY_OFFSET(ipv4.tp);
 869                 icmp_key = nla_data(a[OVS_KEY_ATTR_ICMP]);
 870                 swkey->ipv4.tp.src = htons(icmp_key->icmp_type);
 871                 swkey->ipv4.tp.dst = htons(icmp_key->icmp_code);
 872                 break;
 873         }
 874
 875         return 0;
 876 }
 877
 878 static int ipv6_flow_from_nlattrs(struct sw_flow_key *swkey, int *key_len,
 879                                   const struct nlattr *a[], u32 *attrs)
 880 {
 881         const struct ovs_key_icmpv6 *icmpv6_key;
 882         const struct ovs_key_tcp *tcp_key;
 883         const struct ovs_key_udp *udp_key;
 884
 885         switch (swkey->ip.proto) {
 886         case IPPROTO_TCP:
 887                 if (!(*attrs & (1 << OVS_KEY_ATTR_TCP)))
 888                         return -EINVAL;
 889                 *attrs &= ~(1 << OVS_KEY_ATTR_TCP);
 890
 891                 *key_len = SW_FLOW_KEY_OFFSET(ipv6.tp);
 892                 tcp_key = nla_data(a[OVS_KEY_ATTR_TCP]);
 893                 swkey->ipv6.tp.src = tcp_key->tcp_src;
 894                 swkey->ipv6.tp.dst = tcp_key->tcp_dst;
 895                 break;
 896
 897         case IPPROTO_UDP:
 898                 if (!(*attrs & (1 << OVS_KEY_ATTR_UDP)))
 899                         return -EINVAL;
 900                 *attrs &= ~(1 << OVS_KEY_ATTR_UDP);
 901
 902                 *key_len = SW_FLOW_KEY_OFFSET(ipv6.tp);
 903                 udp_key = nla_data(a[OVS_KEY_ATTR_UDP]);
 904                 swkey->ipv6.tp.src = udp_key->udp_src;
 905                 swkey->ipv6.tp.dst = udp_key->udp_dst;
 906                 break;
 907
 908         case IPPROTO_ICMPV6:
 909                 if (!(*attrs & (1 << OVS_KEY_ATTR_ICMPV6)))
 910                         return -EINVAL;
 911                 *attrs &= ~(1 << OVS_KEY_ATTR_ICMPV6);
 912
 913                 *key_len = SW_FLOW_KEY_OFFSET(ipv6.tp);
 914                 icmpv6_key = nla_data(a[OVS_KEY_ATTR_ICMPV6]);
 915                 swkey->ipv6.tp.src = htons(icmpv6_key->icmpv6_type);
 916                 swkey->ipv6.tp.dst = htons(icmpv6_key->icmpv6_code);
 917
 918                 if (swkey->ipv6.tp.src == htons(NDISC_NEIGHBOUR_SOLICITATION) ||
 919                     swkey->ipv6.tp.src == htons(NDISC_NEIGHBOUR_ADVERTISEMENT)) {
 920                         const struct ovs_key_nd *nd_key;
 921
 922                         if (!(*attrs & (1 << OVS_KEY_ATTR_ND)))
 923                                 return -EINVAL;
 924                         *attrs &= ~(1 << OVS_KEY_ATTR_ND);
 925
 926                         *key_len = SW_FLOW_KEY_OFFSET(ipv6.nd);
 927                         nd_key = nla_data(a[OVS_KEY_ATTR_ND]);
 928                         memcpy(&swkey->ipv6.nd.target, nd_key->nd_target,
 929                                sizeof(swkey->ipv6.nd.target));
 930                         memcpy(swkey->ipv6.nd.sll, nd_key->nd_sll, ETH_ALEN);
 931                         memcpy(swkey->ipv6.nd.tll, nd_key->nd_tll, ETH_ALEN);
 932                 }
 933                 break;
 934         }
 935
 936         return 0;
 937 }
 938
 939 static int parse_flow_nlattrs(const struct nlattr *attr,
 940                               const struct nlattr *a[], u32 *attrsp)
 941 {
 942         const struct nlattr *nla;
 943         u32 attrs;
 944         int rem;
 945
 946         attrs = 0;
 947         nla_for_each_nested(nla, attr, rem) {
 948                 u16 type = nla_type(nla);
 949                 int expected_len;
 950
 951                 if (type > OVS_KEY_ATTR_MAX || attrs & (1 << type))
 952                         return -EINVAL;
 953
 954                 expected_len = ovs_key_lens[type];
 955                 if (nla_len(nla) != expected_len && expected_len != -1)
 956                         return -EINVAL;
 957
 958                 attrs |= 1 << type;
 959                 a[type] = nla;
 960         }
 961         if (rem)
 962                 return -EINVAL;
 963
 964         *attrsp = attrs;
 965         return 0;
 966 }
 967
 968 /**
 969  * ovs_flow_from_nlattrs - parses Netlink attributes into a flow key.
 970  * @swkey: receives the extracted flow key.
 971  * @key_lenp: number of bytes used in @swkey.
 972  * @attr: Netlink attribute holding nested %OVS_KEY_ATTR_* Netlink attribute
 973  * sequence.
 974  */
 975 int ovs_flow_from_nlattrs(struct sw_flow_key *swkey, int *key_lenp,
 976                       const struct nlattr *attr)
 977 {
 978         const struct nlattr *a[OVS_KEY_ATTR_MAX + 1];
 979         const struct ovs_key_ethernet *eth_key;
 980         int key_len;
 981         u32 attrs;
 982         int err;
 983
 984         memset(swkey, 0, sizeof(struct sw_flow_key));
 985         key_len = SW_FLOW_KEY_OFFSET(eth);
 986
 987         err = parse_flow_nlattrs(attr, a, &attrs);
 988         if (err)
 989                 return err;
 990
 991         /* Metadata attributes. */
 992         if (attrs & (1 << OVS_KEY_ATTR_PRIORITY)) {
 993                 swkey->phy.priority = nla_get_u32(a[OVS_KEY_ATTR_PRIORITY]);
 994                 attrs &= ~(1 << OVS_KEY_ATTR_PRIORITY);
 995         }
 996         if (attrs & (1 << OVS_KEY_ATTR_IN_PORT)) {
 997                 u32 in_port = nla_get_u32(a[OVS_KEY_ATTR_IN_PORT]);
 998                 if (in_port >= DP_MAX_PORTS)
 999                         return -EINVAL;
1000                 swkey->phy.in_port = in_port;
1001                 attrs &= ~(1 << OVS_KEY_ATTR_IN_PORT);
1002         } else {
1003                 swkey->phy.in_port = USHRT_MAX;
1004         }
1005
1006         /* Data attributes. */
1007         if (!(attrs & (1 << OVS_KEY_ATTR_ETHERNET)))
1008                 return -EINVAL;
1009         attrs &= ~(1 << OVS_KEY_ATTR_ETHERNET);
1010
1011         eth_key = nla_data(a[OVS_KEY_ATTR_ETHERNET]);
1012         memcpy(swkey->eth.src, eth_key->eth_src, ETH_ALEN);
1013         memcpy(swkey->eth.dst, eth_key->eth_dst, ETH_ALEN);
1014
1015         if (attrs & (1u << OVS_KEY_ATTR_ETHERTYPE) &&
1016             nla_get_be16(a[OVS_KEY_ATTR_ETHERTYPE]) == htons(ETH_P_8021Q)) {
1017                 const struct nlattr *encap;
1018                 __be16 tci;
1019
1020                 if (attrs != ((1 << OVS_KEY_ATTR_VLAN) |
1021                               (1 << OVS_KEY_ATTR_ETHERTYPE) |
1022                               (1 << OVS_KEY_ATTR_ENCAP)))
1023                         return -EINVAL;
1024
1025                 encap = a[OVS_KEY_ATTR_ENCAP];
1026                 tci = nla_get_be16(a[OVS_KEY_ATTR_VLAN]);
1027                 if (tci & htons(VLAN_TAG_PRESENT)) {
1028                         swkey->eth.tci = tci;
1029
1030                         err = parse_flow_nlattrs(encap, a, &attrs);
1031                         if (err)
1032                                 return err;
1033                 } else if (!tci) {
1034                         /* Corner case for truncated 802.1Q header. */
1035                         if (nla_len(encap))
1036                                 return -EINVAL;
1037
1038                         swkey->eth.type = htons(ETH_P_8021Q);
1039                         *key_lenp = key_len;
1040                         return 0;
1041                 } else {
1042                         return -EINVAL;
1043                 }
1044         }
1045
1046         if (attrs & (1 << OVS_KEY_ATTR_ETHERTYPE)) {
1047                 swkey->eth.type = nla_get_be16(a[OVS_KEY_ATTR_ETHERTYPE]);
1048                 if (ntohs(swkey->eth.type) < 1536)
1049                         return -EINVAL;
1050                 attrs &= ~(1 << OVS_KEY_ATTR_ETHERTYPE);
1051         } else {
1052                 swkey->eth.type = htons(ETH_P_802_2);
1053         }
1054
1055         if (swkey->eth.type == htons(ETH_P_IP)) {
1056                 const struct ovs_key_ipv4 *ipv4_key;
1057
1058                 if (!(attrs & (1 << OVS_KEY_ATTR_IPV4)))
1059                         return -EINVAL;
1060                 attrs &= ~(1 << OVS_KEY_ATTR_IPV4);
1061
1062                 key_len = SW_FLOW_KEY_OFFSET(ipv4.addr);
1063                 ipv4_key = nla_data(a[OVS_KEY_ATTR_IPV4]);
1064                 if (ipv4_key->ipv4_frag > OVS_FRAG_TYPE_MAX)
1065                         return -EINVAL;
1066                 swkey->ip.proto = ipv4_key->ipv4_proto;
1067                 swkey->ip.tos = ipv4_key->ipv4_tos;
1068                 swkey->ip.ttl = ipv4_key->ipv4_ttl;
1069                 swkey->ip.frag = ipv4_key->ipv4_frag;
1070                 swkey->ipv4.addr.src = ipv4_key->ipv4_src;
1071                 swkey->ipv4.addr.dst = ipv4_key->ipv4_dst;
1072
1073                 if (swkey->ip.frag != OVS_FRAG_TYPE_LATER) {
1074                         err = ipv4_flow_from_nlattrs(swkey, &key_len, a, &attrs);
1075                         if (err)
1076                                 return err;
1077                 }
1078         } else if (swkey->eth.type == htons(ETH_P_IPV6)) {
1079                 const struct ovs_key_ipv6 *ipv6_key;
1080
1081                 if (!(attrs & (1 << OVS_KEY_ATTR_IPV6)))
1082                         return -EINVAL;
1083                 attrs &= ~(1 << OVS_KEY_ATTR_IPV6);
1084
1085                 key_len = SW_FLOW_KEY_OFFSET(ipv6.label);
1086                 ipv6_key = nla_data(a[OVS_KEY_ATTR_IPV6]);
1087                 if (ipv6_key->ipv6_frag > OVS_FRAG_TYPE_MAX)
1088                         return -EINVAL;
1089                 swkey->ipv6.label = ipv6_key->ipv6_label;
1090                 swkey->ip.proto = ipv6_key->ipv6_proto;
1091                 swkey->ip.tos = ipv6_key->ipv6_tclass;
1092                 swkey->ip.ttl = ipv6_key->ipv6_hlimit;
1093                 swkey->ip.frag = ipv6_key->ipv6_frag;
1094                 memcpy(&swkey->ipv6.addr.src, ipv6_key->ipv6_src,
1095                        sizeof(swkey->ipv6.addr.src));
1096                 memcpy(&swkey->ipv6.addr.dst, ipv6_key->ipv6_dst,
1097                        sizeof(swkey->ipv6.addr.dst));
1098
1099                 if (swkey->ip.frag != OVS_FRAG_TYPE_LATER) {
1100                         err = ipv6_flow_from_nlattrs(swkey, &key_len, a, &attrs);
1101                         if (err)
1102                                 return err;
1103                 }
1104         } else if (swkey->eth.type == htons(ETH_P_ARP)) {
1105                 const struct ovs_key_arp *arp_key;
1106
1107                 if (!(attrs & (1 << OVS_KEY_ATTR_ARP)))
1108                         return -EINVAL;
1109                 attrs &= ~(1 << OVS_KEY_ATTR_ARP);
1110
1111                 key_len = SW_FLOW_KEY_OFFSET(ipv4.arp);
1112                 arp_key = nla_data(a[OVS_KEY_ATTR_ARP]);
1113                 swkey->ipv4.addr.src = arp_key->arp_sip;
1114                 swkey->ipv4.addr.dst = arp_key->arp_tip;
1115                 if (arp_key->arp_op & htons(0xff00))
1116                         return -EINVAL;
1117                 swkey->ip.proto = ntohs(arp_key->arp_op);
1118                 memcpy(swkey->ipv4.arp.sha, arp_key->arp_sha, ETH_ALEN);
1119                 memcpy(swkey->ipv4.arp.tha, arp_key->arp_tha, ETH_ALEN);
1120         }
1121
1122         if (attrs)
1123                 return -EINVAL;
1124         *key_lenp = key_len;
1125
1126         return 0;
1127 }
1128
1129 /**
1130  * ovs_flow_metadata_from_nlattrs - parses Netlink attributes into a flow key.
1131  * @in_port: receives the extracted input port.
1132  * @key: Netlink attribute holding nested %OVS_KEY_ATTR_* Netlink attribute
1133  * sequence.
1134  *
1135  * This parses a series of Netlink attributes that form a flow key, which must
1136  * take the same form accepted by flow_from_nlattrs(), but only enough of it to
1137  * get the metadata, that is, the parts of the flow key that cannot be
1138  * extracted from the packet itself.
1139  */
1140 int ovs_flow_metadata_from_nlattrs(u32 *priority, u16 *in_port,
1141                                const struct nlattr *attr)
1142 {
1143         const struct nlattr *nla;
1144         int rem;
1145
1146         *in_port = USHRT_MAX;
1147         *priority = 0;
1148
1149         nla_for_each_nested(nla, attr, rem) {
1150                 int type = nla_type(nla);
1151
1152                 if (type <= OVS_KEY_ATTR_MAX && ovs_key_lens[type] > 0) {
1153                         if (nla_len(nla) != ovs_key_lens[type])
1154                                 return -EINVAL;
1155
1156                         switch (type) {
1157                         case OVS_KEY_ATTR_PRIORITY:
1158                                 *priority = nla_get_u32(nla);
1159                                 break;
1160
1161                         case OVS_KEY_ATTR_IN_PORT:
1162                                 if (nla_get_u32(nla) >= DP_MAX_PORTS)
1163                                         return -EINVAL;
1164                                 *in_port = nla_get_u32(nla);
1165                                 break;
1166                         }
1167                 }
1168         }
1169         if (rem)
1170                 return -EINVAL;
1171         return 0;
1172 }
1173
1174 int ovs_flow_to_nlattrs(const struct sw_flow_key *swkey, struct sk_buff *skb)
1175 {
1176         struct ovs_key_ethernet *eth_key;
1177         struct nlattr *nla, *encap;
1178
1179         if (swkey->phy.priority &&
1180             nla_put_u32(skb, OVS_KEY_ATTR_PRIORITY, swkey->phy.priority))
1181                 goto nla_put_failure;
1182
1183         if (swkey->phy.in_port != USHRT_MAX &&
1184             nla_put_u32(skb, OVS_KEY_ATTR_IN_PORT, swkey->phy.in_port))
1185                 goto nla_put_failure;
1186
1187         nla = nla_reserve(skb, OVS_KEY_ATTR_ETHERNET, sizeof(*eth_key));
1188         if (!nla)
1189                 goto nla_put_failure;
1190         eth_key = nla_data(nla);
1191         memcpy(eth_key->eth_src, swkey->eth.src, ETH_ALEN);
1192         memcpy(eth_key->eth_dst, swkey->eth.dst, ETH_ALEN);
1193
1194         if (swkey->eth.tci || swkey->eth.type == htons(ETH_P_8021Q)) {
1195                 if (nla_put_be16(skb, OVS_KEY_ATTR_ETHERTYPE, htons(ETH_P_8021Q)) ||
1196                     nla_put_be16(skb, OVS_KEY_ATTR_VLAN, swkey->eth.tci))
1197                         goto nla_put_failure;
1198                 encap = nla_nest_start(skb, OVS_KEY_ATTR_ENCAP);
1199                 if (!swkey->eth.tci)
1200                         goto unencap;
1201         } else {
1202                 encap = NULL;
1203         }
1204
1205         if (swkey->eth.type == htons(ETH_P_802_2))
1206                 goto unencap;
1207
1208         if (nla_put_be16(skb, OVS_KEY_ATTR_ETHERTYPE, swkey->eth.type))
1209                 goto nla_put_failure;
1210
1211         if (swkey->eth.type == htons(ETH_P_IP)) {
1212                 struct ovs_key_ipv4 *ipv4_key;
1213
1214                 nla = nla_reserve(skb, OVS_KEY_ATTR_IPV4, sizeof(*ipv4_key));
1215                 if (!nla)
1216                         goto nla_put_failure;
1217                 ipv4_key = nla_data(nla);
1218                 ipv4_key->ipv4_src = swkey->ipv4.addr.src;
1219                 ipv4_key->ipv4_dst = swkey->ipv4.addr.dst;
1220                 ipv4_key->ipv4_proto = swkey->ip.proto;
1221                 ipv4_key->ipv4_tos = swkey->ip.tos;
1222                 ipv4_key->ipv4_ttl = swkey->ip.ttl;
1223                 ipv4_key->ipv4_frag = swkey->ip.frag;
1224         } else if (swkey->eth.type == htons(ETH_P_IPV6)) {
1225                 struct ovs_key_ipv6 *ipv6_key;
1226
1227                 nla = nla_reserve(skb, OVS_KEY_ATTR_IPV6, sizeof(*ipv6_key));
1228                 if (!nla)
1229                         goto nla_put_failure;
1230                 ipv6_key = nla_data(nla);
1231                 memcpy(ipv6_key->ipv6_src, &swkey->ipv6.addr.src,
1232                                 sizeof(ipv6_key->ipv6_src));
1233                 memcpy(ipv6_key->ipv6_dst, &swkey->ipv6.addr.dst,
1234                                 sizeof(ipv6_key->ipv6_dst));
1235                 ipv6_key->ipv6_label = swkey->ipv6.label;
1236                 ipv6_key->ipv6_proto = swkey->ip.proto;
1237                 ipv6_key->ipv6_tclass = swkey->ip.tos;
1238                 ipv6_key->ipv6_hlimit = swkey->ip.ttl;
1239                 ipv6_key->ipv6_frag = swkey->ip.frag;
1240         } else if (swkey->eth.type == htons(ETH_P_ARP)) {
1241                 struct ovs_key_arp *arp_key;
1242
1243                 nla = nla_reserve(skb, OVS_KEY_ATTR_ARP, sizeof(*arp_key));
1244                 if (!nla)
1245                         goto nla_put_failure;
1246                 arp_key = nla_data(nla);
1247                 memset(arp_key, 0, sizeof(struct ovs_key_arp));
1248                 arp_key->arp_sip = swkey->ipv4.addr.src;
1249                 arp_key->arp_tip = swkey->ipv4.addr.dst;
1250                 arp_key->arp_op = htons(swkey->ip.proto);
1251                 memcpy(arp_key->arp_sha, swkey->ipv4.arp.sha, ETH_ALEN);
1252                 memcpy(arp_key->arp_tha, swkey->ipv4.arp.tha, ETH_ALEN);
1253         }
1254
1255         if ((swkey->eth.type == htons(ETH_P_IP) ||
1256              swkey->eth.type == htons(ETH_P_IPV6)) &&
1257              swkey->ip.frag != OVS_FRAG_TYPE_LATER) {
1258
1259                 if (swkey->ip.proto == IPPROTO_TCP) {
1260                         struct ovs_key_tcp *tcp_key;
1261
1262                         nla = nla_reserve(skb, OVS_KEY_ATTR_TCP, sizeof(*tcp_key));
1263                         if (!nla)
1264                                 goto nla_put_failure;
1265                         tcp_key = nla_data(nla);
1266                         if (swkey->eth.type == htons(ETH_P_IP)) {
1267                                 tcp_key->tcp_src = swkey->ipv4.tp.src;
1268                                 tcp_key->tcp_dst = swkey->ipv4.tp.dst;
1269                         } else if (swkey->eth.type == htons(ETH_P_IPV6)) {
1270                                 tcp_key->tcp_src = swkey->ipv6.tp.src;
1271                                 tcp_key->tcp_dst = swkey->ipv6.tp.dst;
1272                         }
1273                 } else if (swkey->ip.proto == IPPROTO_UDP) {
1274                         struct ovs_key_udp *udp_key;
1275
1276                         nla = nla_reserve(skb, OVS_KEY_ATTR_UDP, sizeof(*udp_key));
1277                         if (!nla)
1278                                 goto nla_put_failure;
1279                         udp_key = nla_data(nla);
1280                         if (swkey->eth.type == htons(ETH_P_IP)) {
1281                                 udp_key->udp_src = swkey->ipv4.tp.src;
1282                                 udp_key->udp_dst = swkey->ipv4.tp.dst;
1283                         } else if (swkey->eth.type == htons(ETH_P_IPV6)) {
1284                                 udp_key->udp_src = swkey->ipv6.tp.src;
1285                                 udp_key->udp_dst = swkey->ipv6.tp.dst;
1286                         }
1287                 } else if (swkey->eth.type == htons(ETH_P_IP) &&
1288                            swkey->ip.proto == IPPROTO_ICMP) {
1289                         struct ovs_key_icmp *icmp_key;
1290
1291                         nla = nla_reserve(skb, OVS_KEY_ATTR_ICMP, sizeof(*icmp_key));
1292                         if (!nla)
1293                                 goto nla_put_failure;
1294                         icmp_key = nla_data(nla);
1295                         icmp_key->icmp_type = ntohs(swkey->ipv4.tp.src);
1296                         icmp_key->icmp_code = ntohs(swkey->ipv4.tp.dst);
1297                 } else if (swkey->eth.type == htons(ETH_P_IPV6) &&
1298                            swkey->ip.proto == IPPROTO_ICMPV6) {
1299                         struct ovs_key_icmpv6 *icmpv6_key;
1300
1301                         nla = nla_reserve(skb, OVS_KEY_ATTR_ICMPV6,
1302                                                 sizeof(*icmpv6_key));
1303                         if (!nla)
1304                                 goto nla_put_failure;
1305                         icmpv6_key = nla_data(nla);
1306                         icmpv6_key->icmpv6_type = ntohs(swkey->ipv6.tp.src);
1307                         icmpv6_key->icmpv6_code = ntohs(swkey->ipv6.tp.dst);
1308
1309                         if (icmpv6_key->icmpv6_type == NDISC_NEIGHBOUR_SOLICITATION ||
1310                             icmpv6_key->icmpv6_type == NDISC_NEIGHBOUR_ADVERTISEMENT) {
1311                                 struct ovs_key_nd *nd_key;
1312
1313                                 nla = nla_reserve(skb, OVS_KEY_ATTR_ND, sizeof(*nd_key));
1314                                 if (!nla)
1315                                         goto nla_put_failure;
1316                                 nd_key = nla_data(nla);
1317                                 memcpy(nd_key->nd_target, &swkey->ipv6.nd.target,
1318                                                         sizeof(nd_key->nd_target));
1319                                 memcpy(nd_key->nd_sll, swkey->ipv6.nd.sll, ETH_ALEN);
1320                                 memcpy(nd_key->nd_tll, swkey->ipv6.nd.tll, ETH_ALEN);
1321                         }
1322                 }
1323         }
1324
1325 unencap:
1326         if (encap)
1327                 nla_nest_end(skb, encap);
1328
1329         return 0;
1330
1331 nla_put_failure:
1332         return -EMSGSIZE;
1333 }
1334
1335 /* Initializes the flow module.
1336  * Returns zero if successful or a negative error code. */
1337 int ovs_flow_init(void)
1338 {
1339         flow_cache = kmem_cache_create("sw_flow", sizeof(struct sw_flow), 0,
1340                                         0, NULL);
1341         if (flow_cache == NULL)
1342                 return -ENOMEM;
1343
1344         return 0;
1345 }
1346
1347 /* Uninitializes the flow module. */
1348 void ovs_flow_exit(void)
1349 {
1350         kmem_cache_destroy(flow_cache);
1351 }